A local agent loop
It plans, calls a tool, checks the result, and repeats, using a model you already run through Ollama, llama.cpp, LM Studio, or any OpenAI-compatible endpoint on your machine.
computer.vodka is a local agent loop you run on your own machine. It talks to a model on localhost, reads the folder you point it at, calls tools, and waits for approval before anything hard to undo. Early, useful, and honest about the limits.
Public beta opens Q4 2027, after we fix the bugs coming out of private trials. Thanks for waiting.
Most agents send your files and prompts to a company's servers. This one is designed to run against a model on your own machine or network instead, so your files stay where you put them.
It plans, calls a tool, checks the result, and repeats, using a model you already run through Ollama, llama.cpp, LM Studio, or any OpenAI-compatible endpoint on your machine.
It only touches the folder you point it at. Its memory is a set of local notes you can open, edit, or delete yourself.
We trim the context and keep the tool set small so a long session does not grind your machine to a halt.
Deleting files, running installs, or anything hard to undo waits for your yes. You see the plan before it runs.
Read, write, search, run shell commands, and browse pages on your own machine. Enough to finish a task, not a long list of extras.
No lengthy setup and no account to babysit. You go from a trial request to a working agent in one folder, in about the time it takes to install anything else.
Open the interface below and tell the agent you want in, or just type your email. It collects what it needs and sends the request for you.
It runs on macOS, Linux, or Windows. Point it at a local model. After the first setup it does not need an account to keep working.
Open a folder, describe what you want, and watch it work. Approve the steps that matter, and keep a record of what it did.
No dashboard, no control panel. Internet, local files, and connected apps are switches right inside the chat. This preview is a demo of the product, so if you want the real thing, just ask the chat below for a trial.
Tap one to see how it lands in the chat. Live connections ship with the trial.
Message that number and the agent picks up your task from there, using whatever files and apps you've turned on here. The first message in any conversation always gets a password check, so a random text to that number can't make it act on your behalf.
The site is a preview. The installed product is a process on your machine: a planner, a tool runner, memory on disk, and a hard stop before anything destructive. It can finish work in one folder. It is not a cloud copilot with your files on someone else's GPU.
Every turn is the same four steps: plan, pick one tool, read the result, then decide whether to continue or stop. Context is trimmed each turn so a long session still fits on a laptop GPU or a strong CPU.
/v1/chat/completions endpoint on localhost or your LAN.Enough to finish a job in a folder, not a kitchen-sink plugin list. Risky calls wait for a yes.
install commands pause until you confirm the plan.stdout/stderr, so a hung command can't lock the session.No hidden vector store you cannot open. Notes live as plain markdown in the same workspace.
.md files you can open, edit, or delete like any other file.Prompts, file contents, shell output, and embeddings stay on the box you run. This page only emails a trial request. The installed agent does not phone home.
Apps in the chat are switches, not a second control panel. Live connections ship with the trial. You can also text the agent once a conversation is unlocked.
This is an early student-built agent, not a finished platform. The list below is the surface area of the installed runtime, not a marketing version of it.
List, search, open, and rewrite files inside one directory you choose. Grep-style search across that tree.
Create markdown, JSON, CSV, Python, or shell files, then show the diff before it stays on disk.
Shell, tests, formatters, and short programs on your machine. Installs and deletes wait for a yes.
Ollama, llama.cpp, LM Studio, or any OpenAI-compatible server on localhost or your LAN.
Every tool call is written to a local transcript so you can see what it touched after the fact.
Once the binary and model are on the machine, files and prompts do not go to our servers.
Output quality is the local model you attach. A 7B or 14B on a laptop is not GPT-class reasoning.
Useful local models need RAM and, ideally, a GPU. Older phones and low-RAM laptops will stall.
Network fetch is off until you turn it on, and even then it is for pages you ask it to open, not a general crawler.
It can mis-scope a folder, skip a file, or write the wrong thing. Review the diff before you approve.
There is no built-in version control. Keep your own copies of anything you cannot afford to lose.
The chat on this page is a preview. It cannot see your disk. The loop starts after you install the trial.
The agent you install does not need a card. From the Q4 2027 public beta, hosted extras are optional and metered by what you actually use that month. No fixed account fee, no send cap, no plan ceiling. Quiet months stay cheap. Heavy months just bill the work. Anything that can run on your machine stays in the local product.
No subscription · ships with the trial
Plan, tool, observe, stop. Read, write, search, shell, and local fetch inside the folder you choose.
Ollama, llama.cpp, LM Studio, or any OpenAI-compatible endpoint on the machine or LAN.
Markdown notes and a plain transcript. You can open or delete them without an account.
Deletes, overwrites, and installs wait for a yes. That gate does not move to the cloud.
No seat, no monthly fee, no phone-home just to keep using the agent on one computer.
Usage billed each month · no fixed fee, no send limit, only what you use
Metered hosted reasoning for the hard turns. Files stay in your workspace unless you attach them.
Sync chat and memory notes across a laptop and a phone. A single disk cannot do that by itself.
A hosted gateway so the agent can pick up SMS or chat when the laptop is closed. Bill by messages handled, not a plan cap.
Gmail, Calendar, Slack, Drive, and similar need an OAuth broker we host. You pay for the calls you make.
A hosted runner keeps a long task going after the lid closes. Local processes die with the machine.
A second person can approve a risky step and read the same transcript. That is a server product.
See tokens, connector calls, and messages for the month. Export a receipt. Stop anytime and local keeps working.
Wake the agent on a timer or an inbound event. Hosted only, because a closed laptop cannot keep a clock.
We design, build, and run computer.vodka on our own machines in Aotearoa. Raman is in Wellington. Jayson, Ashly, and Emily are in Auckland. Each person owns a part of the agent and still reviews the rest.
Owns the agent loop: plan, tool call, observe, stop. Makes local models through Ollama, llama.cpp, and LM Studio feel usable on a laptop, especially Apple Silicon. Also keeps the tool runner and context trimming from grinding a long session to a halt.
Turns raw agent behavior into a task you can hand someone who does not live in a terminal. Writes evals against folders, decides what ships in a trial, and cuts anything that only looks good in a demo. Watches private-trial bugs so the Q4 2027 public beta is actually usable.
Designs the chat, the capability switches, and the approval states. Builds the desktop shell so computer.vodka feels like a modern tool instead of a raw CLI. Also owns the small details: the mark, the trial flow in the preview, and how the product reads on a phone.
Works on local notes, session logs, and the permission rules around write, delete, and install. Makes sure a bad plan cannot walk outside the folder you pointed at. If a step is hard to undo, it waits for a yes before it runs.